fix build error
[platform/upstream/openblas.git] / cpp_thread_test / dgemv_thread_safety.cpp
1 #include <iostream>
2 #include <vector>
3 #include <random>
4 #include <future>
5 #include <omp.h>
6 #include "../cblas.h"
7 #include "cpp_thread_safety_common.h"
8
9 void launch_cblas_dgemv(double* A, double* x, double* y, const blasint randomMatSize){
10         const blasint inc = 1;
11         cblas_dgemv(CblasColMajor, CblasNoTrans, randomMatSize, randomMatSize, 1.0, A, randomMatSize, x, inc, 0.1, y, inc);
12 }
13
14 int main(int argc, char* argv[]){
15         blasint randomMatSize = 1024; //dimension of the random square matrices and vectors being used
16         uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested
17         uint32_t numTestRounds = 16; //number of testing rounds before success exit
18         
19         if (argc > 4){
20                 std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
21                 abort();
22         }
23         if(argc == 4){
24                 std::vector<std::string> cliArgs;
25                 for (int i = 1; i < argc; i++){
26                         cliArgs.push_back(argv[i]);
27                         std::cout<<argv[i]<<std::endl;
28                 }
29                 randomMatSize = std::stoul(cliArgs.at(0));
30                 numConcurrentThreads = std::stoul(cliArgs.at(1));
31                 numTestRounds = std::stoul(cliArgs.at(2));
32         }
33         
34         std::uniform_real_distribution<double> rngdist{-1.0, 1.0};
35         std::vector<std::vector<double>> matBlock(numConcurrentThreads);
36         std::vector<std::vector<double>> vecBlock(numConcurrentThreads*2);
37         std::vector<std::future<void>> futureBlock(numConcurrentThreads);
38         
39         std::cout<<"*----------------------------*\n";
40         std::cout<<"| DGEMV thread safety tester |\n";
41         std::cout<<"*----------------------------*\n";
42         std::cout<<"Size of random matrices and vectors(N=M): "<<randomMatSize<<'\n';
43         std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
44         std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
45         std::cout<<"This test will need "<<((static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*8)+(static_cast<uint64_t>(randomMatSize)*numConcurrentThreads*8*2))/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
46         
47         std::cout<<"Initializing random number generator..."<<std::flush;
48         std::mt19937_64 PRNG = InitPRNG();
49         std::cout<<"done\n";
50         
51         std::cout<<"Preparing to test CBLAS DGEMV thread safety\n";
52         std::cout<<"Allocating matrices..."<<std::flush;
53         for(uint32_t i=0; i<numConcurrentThreads; i++){
54                 matBlock.at(i).resize(randomMatSize*randomMatSize);
55         }
56         std::cout<<"done\n";
57         std::cout<<"Allocating vectors..."<<std::flush;
58         for(uint32_t i=0; i<(numConcurrentThreads*2); i++){
59                 vecBlock.at(i).resize(randomMatSize);
60         }
61         std::cout<<"done\n";
62         //pauser();
63         
64         std::cout<<"Filling matrices with random numbers..."<<std::flush;
65         FillMatrices(matBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads, 1);
66         //PrintMatrices(matBlock, randomMatSize, numConcurrentThreads);
67         std::cout<<"done\n";
68         std::cout<<"Filling vectors with random numbers..."<<std::flush;
69         FillVectors(vecBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads, 2);
70         std::cout<<"done\n";
71         
72         std::cout<<"Testing CBLAS DGEMV thread safety"<<std::endl;
73         omp_set_num_threads(numConcurrentThreads);
74         for(uint32_t R=0; R<numTestRounds; R++){
75                 std::cout<<"DGEMV round #"<<R<<std::endl;
76                 std::cout<<"Launching "<<numConcurrentThreads<<" threads simultaneously using OpenMP..."<<std::flush;
77                 #pragma omp parallel for default(none) shared(futureBlock, matBlock, vecBlock, randomMatSize, numConcurrentThreads)
78                 for(uint32_t i=0; i<numConcurrentThreads; i++){
79                         futureBlock[i] = std::async(std::launch::async, launch_cblas_dgemv, &matBlock[i][0], &vecBlock[i*2][0], &vecBlock[i*2+1][0], randomMatSize);
80                 }
81                 std::cout<<"done\n";
82                 std::cout<<"Waiting for threads to finish..."<<std::flush;
83                 for(uint32_t i=0; i<numConcurrentThreads; i++){
84                         futureBlock[i].get();
85                 }
86                 std::cout<<"done\n";
87                 std::cout<<"Comparing results from different threads..."<<std::flush;
88                 for(uint32_t i=2; i<(numConcurrentThreads*2); i+=2){ //i is the index of vector x, for a given thread
89                         for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize); j++){
90                                 if (std::abs(vecBlock[i+1][j] - vecBlock[1][j]) > 1.0E-13){ //i+1 is the index of vector y, for a given thread
91                                         std::cout<<"ERROR: one of the threads returned a different result! Index : "<<i+1<<std::endl;
92                                         std::cout<<"CBLAS DGEMV thread safety test FAILED!"<<std::endl;
93                                         return -1;
94                                 }
95                         }
96                 }
97                 std::cout<<"OK!\n"<<std::endl;
98         }
99         std::cout<<"CBLAS DGEMV thread safety test PASSED!\n"<<std::endl;
100         return 0;
101 }